Unlock instant, AI-driven research and patent intelligence for your innovation.

Federated learning for anomaly detection

a technology of anomaly detection and learning, applied in the field of anomaly detection, can solve the problems of limiting the availability of such data, and data collected at an edge device may not be forwarded to a central location,

Pending Publication Date: 2022-02-10
NEC LAB AMERICA
View PDF0 Cites 2 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Benefits of technology

The patent describes a method and system for training a neural network model by collecting and aggregating model exemplar information from edge devices. A global model is trained using federated averaging and constrained clustering, and then transmitted to the edge devices. This approach allows for efficient training of neural network models using information from multiple devices, reducing the need for centralized training. The technical effect of this invention is improved training efficiency and flexibility for neural network models.

Problems solved by technology

While machine learning models benefit from being trained on large amounts of data, data sharing policies may limit the availability of such data.
For example, data collected at an edge device may not be forwarded to a central location, out of concern for privacy violations.

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Federated learning for anomaly detection
  • Federated learning for anomaly detection
  • Federated learning for anomaly detection

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0014]Federated learning may be used in a variety of machine learning applications, particularly where security and privacy make training a machine learning model challenging. In particular, federated, unsupervised anomaly detection, which makes use of data collected during normal operation of heterogeneously distributed, isolated edge devices, may take into account unseen heterogeneous normal data at various devices, and may take into account the heterogeneity of local models that are trained on biased data.

[0015]Toward that end, an exemplar-based approach for multivariate time series anomaly detection can preserve data privacy on edge devices and can handle data that is not distributed in an independent, identical way over edge devices. Local exemplars are used to perform anomaly detection and to capture a data distribution of clients, which may then be used to guide federated aggregation of local models in a distribution-aware manner. Each edge device may update relevant exemplar...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

Methods and systems for training a neural network include collecting model exemplar information from edge devices, each model exemplar having been trained using information local to the respective edge devices. The collected model exemplar information is aggregated together using federated averaging. Global model exemplars are trained using federated constrained clustering. The trained global exemplars are transmitted to respective edge devices.

Description

RELATED APPLICATION INFORMATION[0001]This application claims priority to U.S. Provisional Patent Application No. 63 / 062,031, filed on Aug. 6, 2020, to U.S. Provisional Patent Application No. 63 / 070,437, filed on Aug. 26, 2020, and to U.S. Provisional Patent Application No. 63 / 075,450, filed on Sep. 8, 2020, each incorporated herein by reference in its entirety.B ACKGROUNDTechnical Field[0002]The present invention rela85tes to anomaly detection in cyber-physical systems, and, more particularly, to the use of federated learning among local models to improve model efficacy.Description of the Related Art[0003]While machine learning models benefit from being trained on large amounts of data, data sharing policies may limit the availability of such data. For example, data collected at an edge device may not be forwarded to a central location, out of concern for privacy violations.SUMMARY[0004]A method for training a neural network includes collecting model exemplar information from edge d...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Applications(United States)
IPC IPC(8): G06N3/08G06N3/04
CPCG06N3/08G06N3/04G06N3/044
Inventor SONG, DONGJINCHEN, YUNCONGLUMEZANU, CRISTIANMIZOGUCHI, TAKEHIKOCHEN, HAIFENGZHU, WEI
Owner NEC LAB AMERICA